ML-powered crisis prediction — regional forecasts, contagion analysis, and systemic risk assessment
This page runs the dashboard's signal history through a set of forward-looking models and outputs, per region, a predicted crisis score at 24, 48, or 72 hours ahead - along with a confidence level, a baseline / optimistic / pessimistic scenario range, and the key drivers (which features pushed the forecast up or down). It's separate from the Markets page, which shows crowd-sourced odds from Polymarket rather than our internal models.
How to read it. Focus on three things, in order: confidence (below ~65% means the model is unsure - treat it as a flag to look, not to act), direction (is the region worsening or improving relative to now), and the scenario spread (a wide pessimistic-to-optimistic range means the forecast is regime-dependent, not a point estimate). The driver list tells you which features actually moved the prediction - if none of them look credible to you, ignore the number.
Limitations. Forecasts are trained on post-2000 crisis patterns and will under-predict genuinely novel regimes. Sub-regime transitions happen faster than any model trained on historical data can learn. Treat the output as a structured hypothesis about what the visible signals are pointing toward - not a commitment. The AI brief at the top synthesizes all of this into a plain-language daily read.